ape: Remove a couple of posisble infinte loops that can happen with misbehaving hardware. (#154)v0.4.47

- Stop waiting for the RX CPUs to reset after 1 second of waiting.
- Stop waiting for the RMU to reset a bad packet after 100ms.

Closes GH-96.

3 files changed, 39 insertions, 13 deletions

diff --git a/ape/include/ape_main.h b/ape/include/ape_main.h
index 6c2fec0..d0ef27e 100644
--- a/ape/include/ape_main.h
+++ b/ape/include/ape_main.h
@@ -45,6 +45,8 @@
 #ifndef APE_H
 #define APE_H
 
-void initRMU(void);
+void RMU_init(void);
+
+void RMU_resetBadPacket(void);
 
 #endif /* APE_H */
diff --git a/ape/main.c b/ape/main.c
index 740e689..16cca64 100644
--- a/ape/main.c
+++ b/ape/main.c
@@ -64,6 +64,7 @@
 #endif
 
 #define RMU_WATCHDOG_TIMEOUT_MS (10)
+#define RX_CPU_RESET_TIMEOUT_MS (1000) /* Wait up to 1 second for each RX CPU to start */
 
 static NetworkPort_t *gPort;
 
@@ -112,6 +113,8 @@ void handleCommand(void)
 
 void wait_for_rx(const volatile DEVICE_t *device, const volatile SHM_t *shm)
 {
+    uint32_t startTime = Timer_getCurrentTime1KHz();
+
     bool waiting = true;
     do
     {
@@ -126,6 +129,12 @@ void wait_for_rx(const volatile DEVICE_t *device, const volatile SHM_t *shm)
             // Firmware has finished initialization.
             waiting = false;
         }
+
+        if (Timer_didTimeElapsed1KHz(startTime, RX_CPU_RESET_TIMEOUT_MS))
+        {
+            printf("RX CPU reset timeout.\n");
+            waiting = false;
+        }
     } while (waiting);
 }
 
@@ -152,11 +161,7 @@ void handleBMCPacket(void)
         if (stat.bits.Bad)
         {
             // ACK bad packet.
-            APE_PERI.BmcToNcRxControl.bits.ResetBad = 1;
-            while (APE_PERI.BmcToNcRxControl.bits.ResetBad)
-            {
-                // Wait
-            }
+            RMU_resetBadPacket();
         }
         else
         {
@@ -202,7 +207,7 @@ void handleBMCPacket(void)
                         printf("Resetting TX...\n");
                         // Reset, as it's likely locked up now.
                         wait_for_all_rx();
-                        initRMU();
+                        RMU_init();
                         NCSI_reload(AS_NEEDED);
                     }
                 }
@@ -231,7 +236,7 @@ void handleBMCPacket(void)
             if (Timer_didTimeElapsed1KHz(inProgressStartTime, RMU_WATCHDOG_TIMEOUT_MS))
             {
                 printf("RMU Hang detected, resetting.\n");
-                initRMU();
+                RMU_init();
                 packetInProgress = false;
             }
         }
@@ -254,7 +259,7 @@ void checkSupply(bool alwaysReport)
         {
             printf("Powered off, resetting\n");
             wait_for_all_rx();
-            initRMU();
+            RMU_init();
             NCSI_reload(ALWAYS_RESET);
         }
 
@@ -303,13 +308,13 @@ void __attribute__((noreturn)) loaderLoop(void)
             {
                 printf("host started\n");
                 wait_for_all_rx();
-                initRMU();
+                RMU_init();
                 NCSI_reload(NEVER_RESET);
             }
             else
             {
                 wait_for_all_rx();
-                initRMU();
+                RMU_init();
                 if (SHM_HOST_DRIVER_STATE_STATE_UNLOAD == host_state)
                 {
                     printf("host unloaded.\n");
@@ -396,7 +401,7 @@ void __attribute__((noreturn)) __start()
 
     checkSupply(true);
 
-    initRMU();
+    RMU_init();
 
     if (full_init)
     {
diff --git a/ape/rmu.c b/ape/rmu.c
index afc8fb4..fac1e75 100644
--- a/ape/rmu.c
+++ b/ape/rmu.c
@@ -45,9 +45,17 @@
 #include <APE_APE.h>
 #include <APE_APE_PERI.h>
 #include <Network.h>
+#include <Timer.h>
 #include <ape_main.h>
 
-void initRMU(void)
+#ifndef CXX_SIMULATOR
+#include <ape_console.h>
+#include <printf.h>
+#endif
+
+#define RMU_RESET_TIMEOUT_MS (2000) /* Allow up to two seconds for a bad packet to be reset. */
+
+void RMU_init(void)
 {
     RegAPEMode_t mode;
     mode.r32 = APE.Mode.r32;
@@ -102,9 +110,20 @@ void initRMU(void)
     arbControl.bits.TKNREL = 0x14;
     APE_PERI.ArbControl = arbControl;
 
+    RMU_resetBadPacket();
+}
+
+void RMU_resetBadPacket(void)
+{
+    uint32_t startTime = Timer_getCurrentTime1KHz();
+
     APE_PERI.BmcToNcRxControl.bits.ResetBad = 1;
     while (APE_PERI.BmcToNcRxControl.bits.ResetBad)
     {
         // Wait
+        if (Timer_didTimeElapsed1KHz(startTime, RMU_RESET_TIMEOUT_MS))
+        {
+            printf("RMU packet reset failed.\n");
+        }
     }
 }